Why We Built an Open-source Virtual Feature Store? Learnings from Serving 100M Monthly Active Users

Simba Khadder shares his insights during a talk at the MLOps Meetup in San Francisco, highlighting their experience with 100 million monthly active users. They discussed their failures and learnings, primarily focusing on building a recommended system. By utilizing embeddings, they generated candidate song recommendations and refined the rankings. Simba emphasized the significance of resolving workflow challenges to enhance the effectiveness of data scientists. They introduced the concept of a feature store, which revolutionized their approach to data science.